How accurate are our outage estimates
When a country has an internet outage, its page shows how long similar outages lasted: half ended within X hours, 80% within Y hours. A statement like that is only worth something if it holds. Here we check it against what actually happened.
The 80% bound is more cautious than stated: outages end within it more often than promised. The half-way time is a little long: more than half of outages end sooner.
How we check
Each estimate is rebuilt as it could have been given when the outage began: only outages that had already ended by then are used, with the same rules as on the outage page (same country and same pattern if there are at least 5 such cases, otherwise the same pattern anywhere, otherwise all outages). Then we compare it with the real duration. Outages at the very start of our history, when fewer than 3 earlier cases existed, are left out.
By kind of outage
| Kind | Checked | Within 80% bound | Within half-way time |
|---|---|---|---|
| Problem at one operator | 156 | 86% | 56% |
| Minor dip | 56 | 88% | 61% |
| No clear pattern | 26 | 85% | 65% |
| Scheduled shutdown | 19 | 100% | 100% |
| Whole country down at once | 15 | 87% | 67% |
| Failure spreading between providers | 15 | 73% | 60% |
| Other kinds (fewer than 10 each) | 8 | 75% | 63% |
By month
| Month | Checked | Within 80% bound | Within half-way time |
|---|---|---|---|
| November 2026 | 5 | 100% | 60% |
| October 2026 | 84 | 85% | 61% |
| September 2026 | 80 | 89% | 66% |
| August 2026 | 75 | 84% | 57% |
| July 2026 | 51 | 86% | 63% |
Latest checked outages
| Country | Start (UTC) | Estimate, h: half / 80% | Actual, h | Result |
|---|---|---|---|---|
| Eritrea | 5 Nov, 01:40 | 1.2 / 3.2 | 1.0 | within half-way time |
| Vatican City | 4 Nov, 01:40 | 0.8 / 3.3 | 0.5 | within half-way time |
| Chad | 2 Nov, 23:20 | 1.5 / 3.3 | 2.7 | within 80% bound |
| Central African Republic | 2 Nov, 09:30 | 0.5 / 2.0 | 0.3 | within half-way time |
| São Tomé & Príncipe | 1 Nov, 03:00 | 0.7 / 1.0 | 1.0 | within 80% bound |
| São Tomé & Príncipe | 30 Oct, 15:50 | 1.2 / 3.2 | 1.0 | within half-way time |
| Guyana | 30 Oct, 12:20 | 3.8 / 3.8 | 2.8 | within half-way time |
| Mongolia | 30 Oct, 11:20 | 1.5 / 3.3 | 2.8 | within 80% bound |
| São Tomé & Príncipe | 30 Oct, 04:10 | 1.2 / 3.2 | 1.2 | within half-way time |
| Nauru | 29 Oct, 16:20 | 0.5 / 0.7 | 3.2 | longer than estimate |
| Chad | 29 Oct, 12:50 | 1.5 / 3.3 | 2.2 | within 80% bound |
| Cameroon | 29 Oct, 12:50 | 1.2 / 3.2 | 1.3 | within 80% bound |
| São Tomé & Príncipe | 28 Oct, 15:50 | 1.2 / 3.2 | 0.7 | within half-way time |
| Papua New Guinea | 28 Oct, 12:00 | 1.2 / 3.3 | 0.3 | within half-way time |
| Solomon Islands | 28 Oct, 11:10 | 1.5 / 2.7 | 3.3 | longer than estimate |
| Samoa | 28 Oct, 11:10 | 1.2 / 3.3 | 3.2 | within 80% bound |
| Tonga | 28 Oct, 11:10 | 1.2 / 3.3 | 3.0 | within 80% bound |
| U.S. Virgin Islands | 28 Oct, 03:10 | 1.2 / 3.3 | 1.3 | within 80% bound |
| South Sudan | 27 Oct, 05:10 | 1.2 / 3.3 | 0.3 | within half-way time |
| Haiti | 25 Oct, 23:10 | 1.2 / 3.3 | 0.7 | within half-way time |
Estimates recorded at the time
From 5 Nov 2026 we also store the first estimate shown for each outage as it happens, so it can be checked without any reconstruction. Checked so far: 1 of 2 recorded. Too few finished outages yet for a percentage; it appears from 10.